基于监控视频的行人检测与跟踪方法研究  

Research on Pedestrian Detection and Tracking Methods Based on Surveillance Videos

作　　者：张旭[1] 李飞[1] ZHANG Xu;LI Fei(School of Information Science and Engineering,Shenyang University of Technology,Shenyang 110870,China)

机构地区：[1]沈阳工业大学信息科学与工程学院,沈阳110870

出　　处：《微处理机》2025年第1期50-54,共5页Microprocessors

摘　　要：针对监控场景中行人遮挡导致的检测漏检和精度下降问题,以及跟踪算法中精度与计算复杂度难以平衡的挑战,在YOLOv8基础上引入掩码注意力网络,提出M-YOLOv8行人检测算法。同时,通过将M-YOLOv8与优化后的DeepSort算法相结合,并对行人重识别模型进行轻量化处理,构建了一个完整的行人检测与跟踪方案。实验结果表明,改进后的算法在保持较高检测精度的同时,具有较低的计算成本,可有效应用于监控视频中的行人检测与跟踪任务。To address the challenges of missed detections and decreased accuracy caused by pedestrian occlusion in surveillance scenarios,as well as the difficulty in balancing accuracy and computational complexity in tracking algorithms,this paper proposes the M-YOLOv8 pedestrian detection algorithm by introducing a mask attention network based on YOLOv8.Furthermore,by combining M-YOLOv8 with an optimized DeepSort algorithm and implementing a lightweight person re-identification model,a compre-hensive pedestrian detection and tracking solution is developed.Experimental results demonstrate that the improved algorithm maintains high detection accuracy while reducing computational cost,making it effective for pedestrian detection and tracking tasks in surveillance videos.

关 键 词：监控视频 行人检测 行人跟踪 YOLOv8算法 DeepSORT算法 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]